Pydantic DeepAgents
A self-hosted coding assistant and Python agent framework that branches runs to compare approaches and supports local models through Ollama.
Alternative to:
Pydantic DeepAgents is a terminal coding assistant and Python framework for developers who want to use or build their own AI agents. Both share a Pydantic AI foundation for carrying out tasks across files and tools.
- Parallel approaches: Split an active run into branches with separate file changes, review their diffs, and choose a continuation yourself or let an AI judge assess results using test outcomes.
- Coding and delegation: The assistant reads and edits files, runs shell commands, tracks task plans, and delegates focused work to sub-agents.
- Web and integrations: Agents can search the web, automate browsers through Playwright, and connect to MCP servers such as GitHub and Figma.
- Session continuity: Persistent memory carries information between sessions, checkpoints let you save or rewind work, and automatic summaries help long conversations fit within the model's context window.
- Custom agents: Build assistants with typed Python tools, use framework components independently, and add optional Docker sandboxing and approval gates for tool actions.
The project is self-hosted and open source under the MIT license. It supports models with tool-calling capabilities, including Qwen3 and Llama through Ollama, plus Claude, GPT, Gemini and OpenAI-compatible APIs. Local model processing runs on your hardware; cloud model choices send requests to the selected provider. Agent file storage can live in memory or on disk.
